Abstract
During the transition into high school, adolescents sort large sets of unfamiliar peers into prototypical peer crowds thought to share similar values, behaviors, and interests (e.g., Jocks). Often, such sorting is based solely on appearance. This study investigates the accuracy of this sorting process in relation to actual characteristics using video and survey data from a longitudinal sample of U.S. youths who attended high school in the mid- to late-2000s. To simulate this sorting process, we asked same-birth-cohort strangers to view short videos of youths at age 15 and to classify those strangers into likely crowd membership. We then compared the classifications they made to how adolescents characterized themselves at that same time point. Results show that peer crowd classification predicts aspects of unknown peers’ mental health, academic achievement, extracurricular involvement, social status, and risk-taking behaviors.
The transition into high school is a critical social psychological experience in the lives of young people, prompting dramatic shifts in their academic and social environments. In addition to facing higher academic expectations than ever before (McCallumore and Sparapani 2010), adolescents encounter many new peers during a developmental period in which their needs for individuation, belonging, and peer acceptance are at all-time highs (Wentzel, Baker, and Russell 2009). To navigate this complex landscape, adolescents must quickly categorize a sea of new peers into groups of people to befriend, avoid, or approach—often based solely on their appearance. Doing so provides a social roadmap for locating like-minded friends, steering clear of potential foes, and finding classmates that elevate a youth’s own social status (Brechwald and Prinstein 2011).
One way in which adolescents sort groups of unfamiliar peers is by classifying them into peer crowds, or reputation-based clusters of youths stereotyped as having similar values, behaviors, and interests (Brown and Larson 2009). Peer crowds (e.g., Populars) function as prototypes, or abstract representations of groups’ stereotypical features (Cantor and Mischel 1979). Adolescents assign peers to crowds by considering the extent to which their appearances conform to archetypal images associated with these prototypes (Stone and Brown 1999). In doing so, adolescents transform abstract, sociocultural representations of crowds into categories that facilitate peer interaction.
This process, however, is not without error because adolescents may classify peers based on crowd prototypes and associated stereotypes that are out of sync with youths’ actual characteristics. For example, a peer may visibly resemble the Jock prototype but not be on a sports team. The likelihood of adolescents misclassifying peers may be higher during the transition into high school compared to later on in their schooling because adolescents may rely on outward appearance during initial interactions but learn more about their classmates over time. Because the classmates that youth choose to approach or avoid in adolescence shape their academic trajectories (Barber, Eccles, and Stone 2001) as well as their physical and socioemotional well-being (Moran et al. 2017), an important question is how well the processes adolescents use to quickly sort their new classmates into peer crowds predict those peers’ actual characteristics during their first year of high school.
Despite the importance of this question, few studies have examined how adolescents sort unfamiliar classmates into crowds. This lack of attention likely reflects the methodological challenges of this approach, such as the logistics of having adolescents classify large groups of strangers into crowds. In contrast, prior research has focused on having youths self-report their crowd membership or describe the crowds in their own high school. Peer crowd classification by unknown peers—and whether these snap judgments align with classified youths’ actual characteristics—is, thus, an understudied yet important area given the ubiquity of this process during the transition into high school and the cascading ways in which stereotyping based on assumed crowd membership may affect youths’ later well-being.
In this spirit, this study considers peer crowds as informative but imperfect prototypes. It investigates associations between crowd prototypes and youths’ self-reported characteristics using both video and survey data from a longitudinal sample of U.S. youths who attended high school in the mid- to late-2000s. We compared classifications of likely crowd membership made by same-birth-cohort strangers who viewed short videos of adolescents at age 15 to the adolescents’ self-reported traits, behaviors, and activities from the same time point. In doing so, this study addresses methodological gaps that have hindered efforts to understand important social psychological processes during the transition into high school. This study also offers insights into the link between perceptions and reality and the social psychological shortcuts that can, on the one hand, make youths’ lives easier and, on the other hand, pigeonhole others into categories different from their true selves.
Background
Identifying Socioculturally Relevant Peer Crowd Prototypes
During the transition into high school, adolescents tend to sort large sets of unknown peers into consensually recognized categories in order to organize and navigate this new social setting. Among the categories that adolescents use to sort their peers are peer crowd prototypes. Some crowd prototypes are stable across space and time, whereas others ebb and flow, yet all are shaped by contemporary cultural contexts such as broad national trends, popular media representations, and local experiences. These contexts shape which crowd prototypes are called to mind as well as which stereotypes regarding their members’ outward appearances, activities, behaviors, and beliefs are invoked (Crabbe et al. 2019). Adolescents enter high school with knowledge of these socioculturally relevant crowd prototypes and their accompanying stereotypes, which they then apply to new interpersonal interactions.
Researchers have used multiple strategies for investigating crowd prototypes. One strategy relies on using popular culture representations of crowds (Barber et al. 2001), but this ultimately limits prototypes to sensationalized crowds in popular media. Another strategy uses crowd landscapes as described by students who attend a small number of specific schools (Milner 2006), yet this method overgeneralizes unique crowds found in individual schools. These cautions suggest we need a more representative list of crowds and their associated stereotypes generated by adolescents across a variety of schools and geographies (Moran et al. 2017). The Data and Methods section describes how we generated such a list for this study.
Matching Peers to Crowds Based on Visible Cues
Adolescents classify peers into crowds through a cognitive process (typically quick and unconscious) that matches what they know about their peers to what they know about the socioculturally relevant crowd prototypes at that moment in time (Payne, Bettman, and Johnson 1992). Especially when adolescents sort previously unknown classmates in large high schools, classifications may be based solely on visible cues (e.g., dress) rather than on characteristics that are not so easily discernable at first glance (Quadfleig and Macrae 2011). For example, an adolescent sorting an unknown peer into the Fine Arts crowd may not know the peer’s actual arts involvement and thus the classification would depend on visible cues.
Associating Characteristics with Each Peer Crowd
Visible cues may signal additional, nonvisible characteristics of the peer being classified (Macrae and Quadflieg 2010). Indeed, once adolescents are categorized as belonging to a crowd, stereotypes associated with that crowd prototype are applied to the classified youths. In other words, if an individual visibly resembles a crowd prototype, that individual may also be thought to conform to the prototype in other stereotypical ways (Gelman 2004). Thus, classifying a peer into a crowd based on the peer’s visible characteristics provides adolescents with additional information about the person’s possible activities, behaviors, and beliefs without ever having interacted with them.
Classifications based on crowd prototypes may be most effective for facilitating social interactions if the snap judgments they inform reflect the true characteristics of the peers in question. In reality, however, adolescents may not conform to the stereotypes of their perceived crowd. As a result, these classifications may be unhelpful, or worse counterproductive, as adolescents map out their new school relationships.
Summary and Research Question
This study examines the extent to which adolescents’ self-reported traits, behaviors, and activities align with the expected characteristics that their appearance signals to unknown peers based on their perceived crowd membership. By studying youths who attended schools across multiple states and by having classifications of likely crowd membership come from a pool of same-birth-cohort raters who did not know the youths and also attended many different high schools, we illuminated classification processes within a contemporary youth cohort that transcend any one school. We developed and tested hypotheses using a three-step process, detailed in the following. In so doing, this study addresses the overarching question: How well do crowd classifications by unknown others based on visible cues correlate with the classified youths’ self-reported characteristics?
Methods
This study used a subsample of children in the NICHD Study of Early Child Care and Youth Development (SECCYD). Children were recruited at birth in 1991 from hospitals in 10 sites across the United States, resulting in an initial sample of 1,346 children and families. Although this sample was not nationally representative, participating families reflected the demographic characteristics of the people living in the communities surrounding the participating hospitals. Of the original 1,346 children, 865 were followed up to age 15, with demographic, socioemotional, academic, and visual information gathered at 11 time points across their lives. The 865 youths were statistically equivalent to the 499 who left the study before age 15 in terms of race/ethnicity and family income-to-needs ratio but were significantly more female, had mothers that had completed more education, and were more evenly distributed across sites. 1
We used SECCYD data collected from surveys, transcripts, and videotapes featuring a discussion between the study children and a parent during the 15-year data collection wave. Our research team took seven-second clips from these videotaped interactions that showed the child’s whole face and body and included visible cues like makeup and clothing that adolescents would use to sort unknown peers into crowds. We edited out other visual and auditory cues that may have signaled peer crowd but would not have been apparent based on a quick glance.
Hypothesis testing followed a three-step approach of determining socioculturally relevant peer crowds, classifying unknown peers into crowds, and evaluating peer crowd hypotheses using individual outcomes. Each step used a different set of same-age peers who were unaffiliated with the SECCYD youths.
Determining Socioculturally Relevant Peer Crowds
Focus groups identified peer crowd prototypes around the time period in which SECCYD youths transitioned into high school. Specifically, ten focus groups comprised of 61 recent high school graduates—approximately from the same birth cohort as study youths—were recruited from sociology, kinesiology, and engineering courses at two public universities in 2016 to describe crowds they encountered in high school. 2
During these focus groups, participants individually brainstormed crowds present in their high schools and wrote each crowd name on a post-it note. A moderator then placed the post-it notes on a whiteboard, asking participants to describe each crowd. Participants next grouped similar crowds, distilling to between five and ten crowds found across schools. Participants were then asked to hypothesize crowd characteristics. The discussion began with the prompt, “We want you to think about the typical characteristics of students in this group. What do they look like? What do they like to do?,” and then worked through questions regarding mental health, academics, extracurricular activities, social status, and risk-taking.
Audio transcriptions from each focus group were analyzed using MAXQDA software by the first author and two research assistants to identify common crowds across focus groups and their associated characteristics. Overall, focus groups identified nine crowds used in the current study—namely, Populars, Jocks, Smarts, Fine Arts, Druggie/Stoners, Emo/Goths, Anime/Mangas, Troublemakers, and Loners. Characteristics of each crowd that emerged from focus groups—and guided subsequent analyses—were summarized. 3 Crowds were ranked as having low, middle, or high levels of mental health, academic achievement, specific extracurricular participation, social status, and risk-taking. Results from this exercise were distilled to produce a set of testable hypotheses, listed in Table 1.
Hypotheses about Each Crowd Based on Focus Groups
We tested whether students from crowds classified as low on a certain construct indeed averaged less of the characteristics of that construct than those from the middle and high groups. Similarly, we tested whether students from crowds classified as part of the middle group on a construct averaged less of that construct than those from the high group.
Classifying Unknown Peers into Crowds
Next, 61 trained undergraduates from two large universities—different from focus group participants but also from approximately the same birth cohort as the SECCYD youths—rated youths’ likely crowd membership. 4 To do so, they logged into an online interface (REDCap) where they were presented with random subsets of video clips of the SECCYD youths from the 15-year data collection wave. We refer to these raters as unknown peers because they did not know or have contact with the SECCYD youths but transitioned to high school during the same historical time. All ratings were completed in 2017–2018.
We used the nine crowds that emerged from the focus groups to prime a similar conception of peer crowd prototypes across raters. For the priming process, we had a separate set of undergraduate research assistants identify internet stock images of males and females that exemplified each of the nine peer crowd descriptions. 5 We paired these images with the written descriptions and presented them to raters every time they logged in to complete their rating tasks. Raters practiced rating several stock images in a group training session, discussing why they chose specific crowd ratings for each image. Raters then were given a week to log into the online interface and rate a random subset of the stock images before rating SECCYD youths.
The rating prompt asked raters, “To what extent would you characterize the adolescent as a member of each of the following groups based on appearance?” They could choose from four responses (not at all, a little, somewhat, completely) for each of the nine crowds. 6
We took several steps to reduce random and systematic error. In relation to random error, we increased the signal-to-noise ratio by having 20 research assistants rate the youths. The resulting 20-rater averages had good reliability. 7 We reduced systematic error through randomization. Specifically, we first randomly assigned raters to rating groups. Each rating group’s members rated the same random set of videos with order randomized within each group so that raters within the same rating group would see the same videos but in a different order. Even with randomization, averages could be systematically higher or lower than true scores to the extent a rating group contained a set of harsher raters. To address this possibility, we had all research assistants rate a subset of the same 30 SECCYD videos to estimate each rater’s average harshness. We then calculated adjusted average ratings for each SECCYD youths by subtracting rater means from individual scores before calculating youths’ average scores. Each youth’s crowd was determined as the crowd for which the youth’s adjusted average score was the highest.
Evaluating Peer Crowd Hypotheses Using Individual Outcomes
We used what the SECCYD youths said about themselves during the 15-year wave in order to test focus-group-generated hypotheses about crowd characteristics. In the sections that follow, we motivate the selected variables in relation to the focus group hypotheses, operationalize these variables, and then provide descriptive statistics for the analysis sample in Table 2. Items comprising each variable can be found in the Online Supplement. 8
Descriptive Statistics for the Analytical Sample
Note: N = 865. GPA = grade point average.
Mental health
Focus groups identified poor mental health, especially depression and loneliness, as hallmarks of the Emo/Goth and Loner crowds. All other crowds were hypothesized as having middle levels of mental health. SECCYD youths had an overall mean depressive score of 1.98 (SD = 2.64) on the Children’s Depressive Inventory Short Form and a loneliness score of 26.27 (SD = 8.76) on the Loneliness and Social Dissatisfaction Questionnaire, with higher scores indicating greater psychological distress (Asher, Hymel, and Renshaw 1984; Kovacs 1992).
Academic achievement
Based on focus group discussions, high achievement was seen as a defining feature of Smarts, whereas members of the Troublemaker, Druggie/Stoner, and Loner crowds were thought to fall at the lower end. All remaining crowds were hypothesized as having middle levels of achievement. Academic achievement was measured by SECCYD youths’ advanced math and English enrollment and unweighted grade point average (GPA) in ninth grade. The youths had a mean overall GPA of 2.98 (SD = .71). A quarter were enrolled in advanced English classes, and more than a third were enrolled in advanced math courses.
Extracurricular involvement
Specific extracurricular involvement was seen by focus groups as a defining characteristic of three crowds: sports for Jocks, arts for Fine Arts, and academic clubs for Smarts. Apart from these specific activities, these crowds—along with the Populars—were considered middle-range extracurricular participants. All remaining crowds were expected to have low extracurricular participation. Extracurricular involvement was assessed by asking youths to indicate the activities in which they had participated during the past year. Overall, 78 percent of adolescents within the analytic sample reported having participated in sports, 48 percent in arts, 17 percent in academic clubs, and 20 percent in nonacademic clubs.
Social status
Focus group participants saw high status as a defining feature of the Populars, Jocks, and Smarts. Conversely, low status was considered a hallmark of the Anime/Manga crowd. All other crowds were hypothesized as having middle levels of status. We examined social status as both perceived likeability and popularity (Sandstrom and Cillessen 2006). SECCYD youths averaged more well-liked (M = 5.72, SD = 1.16) than popular (M = 4.53, SD = 1.83) self-perceptions.
Risk-taking
Focus group participants saw risk-taking as a hallmark of the Troublemakers, Druggie/Stoners, Populars, and Jocks. All other crowds were thought to engage in middle levels of risk-taking. To measure risk-taking, SECCYD youths indicated their engagement in a list of 55 risks (Conger and Elder 1994). Study youths indicated taking part in a relatively low number of risks (M = 6.14), although there was substantial variation within the sample (SD = 5.63).
Covariates
To account for factors that might correlate with crowd membership, models were adjusted for youths’ race/ethnicity, sex/gender, maternal education at childbirth, family income-to-needs ratio at age 15, and data collection site. 9
Plan of Analysis
Missing data
The focal peer crowd variable had no item-level missingness because we restricted the sample to SECCYD youths who had video data from the 15-year data collection wave and, thus, peer crowd ratings. The majority of dependent variables and covariates had some missingness, albeit low (see Table 2). Missingness was more substantial for academic achievement variables taken from transcripts that had been requested by the SECCYD team directly from the schools that the youths had attended.
Prior to analyses, item-level missing data were sequentially imputed using chained equations within the mi impute suite of commands in Stata 15 (StataCorp 2017). This process generated 50 replicate data sets. All variables were included and served as predictors during imputation, meaning that the prediction of each outcome variable included all other dependent variables as auxiliary variables. All item-level missing values were successfully recovered, resulting in all 865 adolescents being included in every model. Sensitivity analyses compared results from models with imputed versus nonimputed outcomes. 10 Because of high levels of consistency between results regardless of the use of imputed or nonimputed data, we present results using fully multiply imputed data. Doing so provides a consistent sample size across models and the greatest precision of estimation, even including the smaller crowds.
Imputation assumes missing at random. Although we cannot definitively evaluate this assumption, especially for the transcript data which had the highest degree of missingness, we compared the covariate characteristics of those with and without transcript variables. 11 SECCYD youths that were missing transcript data did not significantly differ from those with transcript data in terms of sex/gender, race/ethnicity, maternal education, or family income-to-needs ratio but did differ by site, with two sites having significantly more youths with missing transcript data than the other eight. Because principals had to consent to releasing this information to the SECCYD team, missing transcript data likely is a function of factors outside of the youths’ control.
Regression models
We regressed each outcome on dummy indicators of groups of crowds hypothesized by focus group members as having low, middle, or high levels of each outcome and the covariates. We used the mi impute prefix in Stata 15 (StataCorp 2017) to combine estimates from the 50 replicates using Rubin’s rules (Rubin 1987). We tested contrasts between pairs of groups hypothesized as having different levels of each outcome (low vs. high, low vs. middle, middle vs. high). Because hypotheses were directional, we used one-tailed hypothesis tests. For some outcomes, focus group participants placed crowds into only two groups (e.g., low and middle). In these cases, a single contrast was tested.
We also presented overall F tests to jointly determine whether any differences were statistically evident among low, middle, and high groups of crowds. We based conclusions regarding statistical significance on an alpha value of .01.
Predicted means (
Predicted Means for Mental Health Outcomes by Focus-Group-Hypothesized Levels
Predicted Means and Probabilities for Academic Outcomes by Focus-Group-Hypothesized Levels
Predicted Probabilities for Extracurricular Outcomes by Focus-Group-Hypothesized Levels
Predicted Means for Social Status Outcomes by Focus-Group-Hypothesized Levels
Predicted Means for Risk-Taking by Focus-Group-Hypothesized Levels
To indicate whether and where low, middle, and high groups significantly differed from one another, see the pairs of letters within Figures 1 through 5. Because hypothesis tests were based on precise formulas for standard errors of the difference in means or difference in proportions and because all tests were directional, conclusions may differ from overlapping confidence intervals shown by standard error bars in the figures (Mitchell 2012).
When discussing significant contrasts, we present differences in predicted means (Δ
Results
Distribution of Crowds
Unknown peers placed SECCYD youths into all nine peer crowds, although some more than others (see Table 2). Half of the youths were classified as either Populars (28 percent) or Smarts (21 percent). A smaller yet sizable proportion were classified as Jocks (14 percent), Fine Arts (11 percent), or Loners (11 percent). About 5 percent were classified as Druggie/Stoners, and about 3 percent each were classified as Troublemakers, Anime/Mangas, and Emo/Goths.
Overall Tests
We began with overall tests, which showed that focus group assessments of crowds’ characteristics (i.e., low, middle, high) were predictors of depression (F = 10.39, p = .001), loneliness (F = 8.66, p = .003), GPA (F = 5.39, p = .005), sports participation (F = 5.37, p = .005), popularity (F = 6.77, p = .001), likeability (F = 5.00, p = .007), and risk-taking (F = 14.73, p < .001). 16 The overall test approached significance for two additional outcomes: advanced English course-taking (F = 2.50, p = .082) and nonacademic club participation (F = 4.04, p = .044). It was nonsignificant for advanced math course-taking (F = .57, p = .56), arts participation (F = 1.69, p = .184), and academic club participation (F = 1.46, p = .233).
In addition to hypothesized crowd level, covariates were often significant predictors of outcomes. Specifically, being female was positively associated with depressive symptomatology, GPA, and arts participation but negatively associated with risk-taking. Identifying as African American was negatively associated with GPA, advanced math course-taking, and self-perceived popularity and likeability but positively associated with risk-taking. Children with mothers who held a bachelor’s degree at childbirth had significantly higher GPAs and greater enrollment in advanced courses as well as generally higher extracurricular participation than their peers. Site was also a significant predictor of advanced course-taking and popularity. Notably, unknown peers’ assessments of the youths’ hypothesized crowd level were the only significant predictors of youths’ loneliness, with no significant covariate coefficients.
Pairwise Contrasts
We now turn to the pairwise contrasts, discussing them by outcome. We first present contrasts among the three groups of crowds—hypothesized low, middle, and high—that are depicted in Figures 1 through 5. 17 We then describe results from pairwise comparisons among all nine crowds to pinpoint which specific crowds might have driven the three-group results. 18 We additionally provide in the Online Supplement a summary for each construct regarding which crowds conformed or did not conform to focus group hypotheses. 19
Mental health
Recall that focus groups identified low mental health as hallmarks of the Emo/Goth and Loner crowds, with all other crowds expected to have middle levels of mental health.
As hypothesized, the low mental health group had higher average levels of depressive symptoms and loneliness than the group of crowds hypothesized as having middle mental health, with effect sizes of about one-third of a standard deviation (Δ
Turning to pairwise contrasts among the nine crowds, we first confirmed that the two crowds that comprised the low mental health group had statistically similar levels of depressive symptomatology and loneliness to each other and that the remaining seven crowds classified as middle mental health groups were generally similar to one another. 20
Several differences were also evident between crowds expected to have low versus middle mental health. Significant and meaningful differences (effect sizes between .35 and .45) in terms of mental health were evident between Loners compared to Populars and Jocks in depressive symptomatology (Δ
Of note, Smarts also reported more loneliness than Populars (Δ
Academic achievement
Focus groups expected high achievement to be a hallmark of the Smarts, and they also considered low achievement to be a key characteristic of the Loners, Troublemakers, and Druggie/Stoners. All other crowds were hypothesized as having middle levels of achievement.
Lending support to hypotheses, the expected low achievement group had significantly lower GPAs (Δ
When we considered contrasts among the nine crowds, differences between the high (Smarts) and low achievement groups were driven by the Loners and Druggie/Stoners regarding GPA (Δ
Contrary to focus group expectations, pairwise contrasts suggested that the Fine Arts were mistakenly classified as middle rather than high achievers. This misclassification may explain insignificant differences between the high and middle achievement groups in terms of English course-taking. Specifically, Fine Arts were more enrolled in advanced English courses than four crowds, including the low achieving Loners (Δ
Results from pairwise contrasts also suggested that the Anime/Mangas would be better classified as a low as opposed to middle achieving crowd. In addition to being less enrolled in advanced English compared to the aforementioned Fine Arts, Anime/Mangas were significantly less enrolled compared to the Smarts (Δ
Extracurricular involvement
Recall that focus groups expected sports participation to be a defining feature of the Jocks, arts for the Fine Arts, and academic clubs for the Smarts. Focus groups also expected all other groups—besides Populars placed in the middle—to be uninvolved in extracurricular activities.
Consistent with expectations, the high sports participation group—comprised solely of the Jocks—significantly exceeded both the low and middle sports participation groups (Δ
Interestingly, the hypothesized low and middle sports participation groups did not differ from one another in this respect. This could be attributed to several crowds being misclassified by focus group participants as low as opposed to middle sports participants in relation to the SECCYD data. Pairwise comparisons suggested that among the group hypothesized as having low sports participation, the Loners were the only crowd that conformed to this expectation. Specifically, Loners had lower sports participation than three crowds: the Jocks, Populars, and Troublemakers (Δ
Similarly, Smarts, who the focus groups thought would have middle levels of sports participation, were less involved than several crowds—the aforementioned Jocks—but also Populars (Δ
In terms of arts participation, there were no significant differences between groups of crowds hypothesized as having low, middle, or high involvement. After probing crowd-specific pairwise contrasts, however, the Jocks were significantly underinvolved in arts compared to five other crowds, namely, the hypothesized high Fine Arts (Δ
Recall that focus group participants hypothesized that Smarts would have high levels of involvement in academic clubs; Populars, Jocks, and Fine Arts middle levels of involvement; and the remaining crowds low levels. Despite no significant differences between high, middle, and low groups of crowds on these outcomes, it was noteworthy that one specific crowd—the Druggie/Stoners—stood out as having lower academic club participation compared to their peers—namely, the hypothesized high Smarts (Δ
Last, focus group participants hypothesized that Populars, Fine Arts, Jocks, and Smarts would have middle level participation in other types of clubs and that the remaining crowds would have low level participation in those clubs. Results suggested that there were no significant differences between low and middle groups on this outcome (Δ
Social status
Focus group participants hypothesized that Jocks, Populars, and Smarts would have high levels of status; Anime/Mangas would have low; and the remaining crowds would have middle.
We found that the high status group indeed reported significantly higher popularity than the middle status group (Δ
Pairwise contrasts between the nine specific crowds suggested that Jocks and Populars—but not Smarts—drove differences in popularity between the middle and high groups. Populars and Jocks reported higher popularity than most other crowds, including Smarts (Δ
Regarding status, the Loners and Fine Arts also seemed misplaced by the focus groups. Loners and Fine Arts were better suited for the low as opposed to middle status group given their low perceived popularity compared to not only the high status Populars (Δ
When considering the self-perceived likeability of specific crowds, Populars and Anime/Mangas seemed to drive significant differences between the high and low status groups. Populars—but not Jocks and Smarts—considered themselves more likeable than four other crowds (Smarts: Δ
Risk-taking
Focus groups highlighted high levels of risk-taking as a hallmark of the Populars, Jocks, Troublemakers, and Druggie/Stoners, with the remaining crowds hypothesized as comprising a set of middle risk-takers. As expected, we found that—as a group—crowds hypothesized as being high risk-takers had statistically higher levels of self-reported risk-taking compared to the hypothesized middle group of crowds (Δ
When probing pairwise contrasts between crowds, we found that the four crowds that comprised the hypothesized high risk-taking group were statistically similar in their high average risk scores.
22
The remaining five crowds that made up the hypothesized middle risk-taking group generally had similar midrange risk scores to one another, save for the Emo/Goths who significantly exceeded the Smarts (Δ
Pairwise contrasts also suggested that the Smarts and Anime/Mangas were misclassified by focus group members as middle risk-takers when they would more accurately be described as low risk-takers. Specifically, the seemingly risk-averse Smarts had significantly lower levels of risk-taking than all crowds hypothesized as having high levels of risk-taking (Troublemakers: Δ
Discussion
This study investigated the accuracy of the classification of youths by unknown peers into crowds by comparing these classifications to youths’ self-reported characteristics. Our design mimicked the process by which adolescents call up crowd prototypes and assess whether a new classmate’s visible cues match them. Ratings based on a 7-second video clip were compared to what the depicted youths actually said about themselves. In many cases, these associations were consistent with expectations based on what same-age focus groups predicted would be the hallmarks of various crowds.
Summary of Main Findings
Particularly consistent with hypotheses were results for mental health, sports participation, and risk-taking, where focus group expectations were met in the majority of instances. For each outcome, we placed crowds into sets predicted by the focus group members to fall into low, middle, and high levels. The hypothesized low mental health group indeed reported more depressive symptoms and loneliness than the middle group, as predicted. In terms of sports participation, the hypothesized high group was more involved in sports than both the middle and low groups. Last, the hypothesized high risk-taking group reported engaging in a higher average number of risks compared to the middle group. These results suggest that unknown peers classified youths into crowds whose defining features were consistent with what same-age peers had predicted and with the youths’ self-reported mental health, sports involvement, and risk-taking.
Although not uniformly supported, focus group predictions for academic achievement and social status were also met in the majority of instances. For example, being classified in a crowd expected to have high achievement was linked to higher GPAs and advanced English course-taking—but not math—compared to crowds expected to comprise low achievers. Hypothesized middle achievers, however, did not academically differ from the high and low sets of crowds. Similarly, the hypothesized high status set of crowds reported greater popularity than the middle set and greater likeability than the low set but did not otherwise differ from their peers. These results highlight the accuracy with which focus groups identified crowds signaling high and low levels of achievement and status at first glance based on crowd prototypes. At the same time, focus groups less often placed crowds between these extremes in ways consistent with SECCYD youth self-reports.
The only construct that did not conform to any focus group expectations was nonsports extracurricular participation. Specifically, sets of crowds hypothesized as having high arts and academic club participation were not more likely to be involved in these activities than the supposedly middle or low crowds. These results suggest that focus groups’ expectations were not informative about SECCYD youths’ arts, academic, or nonacademic club participation.
After probing contrasts between specific crowds, additional noteworthy patterns emerged that went beyond what focus groups considered crowd hallmarks but are consistent with broader literatures.
First, focus groups did not consider Smarts to be part of the low mental health group despite Smarts reporting greater loneliness than other crowds. This result aligned with work that found youths labeled academically “gifted” experience social isolation because they are seen—and see themselves—as different from other children (Rimm 2000).
Second, both the Fine Arts and Anime/Mangas were hypothesized by focus groups as having middle levels of achievement. Fine Arts, however, had relatively high achievement in terms of their GPAs and advanced English enrollment. Anime/Mangas had high GPAs but were among the crowds with the lowest enrollment in advanced English courses. High achievement among Fine Arts was consistent with documented associations between arts involvement and higher GPAs (DiMaggio 1982). In terms of the Anime/Mangas, work by Chandler-Olcott (2015) suggested that youths in this crowd may feel alienated from English curricula that focus on traditional print literature. They may, therefore, opt out of advanced English courses in exchange for classes that encourage their multimodal skill sets.
Third, Smarts had unexpectedly low sports participation compared to other crowds even though focus group members viewed them as all-around extracurricularly engaged. Past literature separated academic crowds into the well-rounded, highly regarded Scholars and the hyperacademic, socially inept Nerds (Crabbe et al. 2019). In our study, focus group descriptions of the well-rounded Smarts most closely resembled the Scholars of the past. The scholar-like Smarts may be hard to distinguish from other crowds because their well-roundedness may visibly signal characteristics of multiple crowds (e.g., not only academic but also athletic and musical). This may have resulted in some Smarts being classified as Jocks or Fine Arts members in our study. Consequently, the youths that were ultimately classified by unknown peers as Smarts likely better resembled the Nerd prototype. This may also help explain why Smarts reported relatively low status and more loneliness than other crowds despite focus groups considering Smarts to be popular and well liked.
Fourth, despite being placed by focus groups in the middle, Jocks had lower arts participation than almost all other crowds. Pascoe (2003) described how Jocks, especially male Jocks, often reject the arts when they are considered “feminine” for fear of having their masculinity and/or sexuality questioned. Our results supported this notion. When these social constructions persist, arts activities can be stigmatized, and those that participate in them can experience lower status than their peers (Lehman and Dumais 2017). We see this reflected in the low status of Fine Arts members compared to other crowd members, something focus groups had not anticipated.
Last, Smarts and Anime/Mangas engaged in less risk-taking than other crowd members even though focus groups expected them to fall in the middle. Past literature has suggested that status and risk-taking are often positively associated with one another (Sweeting and Hunt 2015). One explanation for this link is the role social events like parties play in presenting youths with opportunities for engaging in risks, like drinking alcohol and hooking up (Gallupe and Bouchard 2013). Smarts and Anime/Mangas may not, as a function of their status, gain entrée to such events and thus may not be exposed to the same risk-taking opportunities as their higher status peers. This may help explain both Anime/Mangas’ and Smarts’ relatively low status and risk-taking.
Future Directions for Research on Adolescence
The results of this study suggest several areas for discussion in social psychology, including both the sociological and psychological fields and their intersection.
The first concerns the need for continuously updating crowd prototypes. One of the strengths of our approach was having focus groups of contemporary youths from the same birth cohort elucidate the current peer crowd landscape that they themselves had experienced. Although focus groups were able to name and describe crowds and thus contribute to and update the canon of literature on crowds, there were several noteworthy differences between classified youths that were not brought up in focus group discussions. In several cases, focus group members misclassified crowds as being part of the low, middle, or high groups relative to the SECCYD sample. For example, despite focus group participants naming several extracurricular crowds as part of the modern high school landscape, our results failed to support the idea that certain crowds were more likely than others to be involved in arts, academic, or nonacademic clubs. One potential explanation is that college admissions requirements favoring well-rounded applicants may encourage broader participation and as a result, reduce the salience of extracurricular-focused crowds (Crabbe et al. 2019). This possibility reinforces the need to update crowd prototypes as crowd traits change or otherwise become obsolete in the face of demographic change, such as larger numbers of students with college aspirations and shifting college admissions requirements.
The second points to the influence past crowds still have on understandings of the current peer crowd landscape. Although our results support updating crowd prototypes, it is also important to note that youths in our study seemed to sometimes draw on past prototypes when they were sorting their peers. In our results, we see that youths identified as Smarts looked like Nerds of the past as opposed to the well-rounded Scholars described by focus groups. Despite being prompted with focus groups’ descriptions of the Smarts, unknown raters found the Nerd prototype of the past difficult to ignore when sorting peers. Although several crowds may no longer be part of the current peer crowd landscape, they should not be rendered unimportant because they may continue to shape youths’ perceptions of peers they do not yet know.
The third concerns the utility of peer crowd sorting in adolescence research and intervention. When implementing large-scale behavior interventions to youths, public health practitioners have leveraged knowledge about the ways that adolescents sort unknown peers into crowds. For instance, antismoking campaigns that target youths in various crowds with tailored imagery have been shown to be more effective at reducing problem behaviors than general health campaigns (Moran et al. 2017). The effectiveness of these programs, however, depends on who is doing the sorting and how efficiently they can sort youths into crowds so that they can receive crowd-specific interventions. Our results suggest that practitioners could rely on unknown peers to do this sorting, especially for interventions related to mental health, sports—but not other extracurricular—activities, and risk-taking. At the same time, practitioners delivering interventions should be wary about publicly labeling youths given within-crowd heterogeneity in behaviors and attitudes and the potential for youths to internalize a certain label that can negatively shape their future behaviors (Adams et al. 2003).
The fourth line of discussion concerns the importance of triangulating results using different operationalizations of constructs. We can contextualize differences in how well adolescents’ classifications of unknown peers predict characteristics of crowd members in terms of how each construct is operationalized. For instance, our most robust findings were based on standardized measures of depression and loneliness and transcript-coded GPAs, each of which was well aligned to the underlying construct of interest and had good variability. Extracurricular activities, on the other hand, might be better measured in the context of how important activities are to the youths doing them. Additionally, Smarts might be found to have more status if we measured others’—as opposed to their own—perceptions (McElhaney, Antonishak, and Allen 2008). Thus, future work can build on this study by differently operationalizing our focal constructs to further elucidate what information an individuals’ physical appearance signals to unknown peers.
A fifth line of discussion concerns the innovativeness of our research design. In addition to contributing insights about how well crowd prototypes work or do not work in predicting the characteristics of unknown peers, our study has several noteworthy features. Specifically, this study uses both visual and survey data from a large, geographically diverse, and contemporary sample of youths; includes a list of crowd prototypes generated by same-age peers; and simulates the transition into high school by presenting visual data to unknown, same-age peers that then classify youths into crowds based only on what they see.
Despite the ingenuity of our mixed-methods study design and applicability of the findings to peer crowd scholars and public health practitioners, this study has several important limitations. First, youths in our sample were disproportionately classified into Populars, Jocks, Smarts, Fine Arts, and Loners, with fewer classified as Emo/Goths, Anime/Mangas, Troublemakers, and Druggie/Stoners. Unequal distributions across individual crowds may result in insufficient power to detect differences relative to smaller crowds. Future work should utilize larger samples or oversample these smaller crowds. Second, the “unknown peers” that defined crowds and classified the sample youths into crowds had strengths—having attended high school at the same time as the sample youths—but also weaknesses—were all current college students. College students may have been overrepresented in certain crowds when they were in high school, leading to ingroup/outgroup homogeneity bias (Rubin and Badea 2007). As a result, our unknown peers may have more nuanced understandings of what college-bound youths are like but less familiarity with the characteristics of noncollege-bound youths.
This potential bias is perhaps most evident in disconnects between focus group expectations about Smarts and real-lived experience. For example, Smarts—who are likely from the same pool of youths that participated in our focus groups—were believed to be popular and well liked by focus groups. Youths classified by unknown peers as Smarts, however, reported among the lowest status in the sample. Future work should therefore investigate whether focus groups and raters from other social locations name and describe the same or additional crowd prototypes and classify SECCYD adolescents differently, something that will be possible using a planned archive of the SECCYD visual images.
In sum, this study employs an innovative, multistep design to elucidate how well the processes youths use to quickly sort unknown peers and navigate a new high school social environment reliably predict their peers’ characteristics. Results suggest prototypes are helpful in accurately assessing adolescents’ mental health, sports participation, risk-taking, academic achievement, and social status—especially at the low and high extremes. They are, however, less predictive of other extracurricular participation, as measured in this study. Results from this study highlight new ways in which peer crowds can be harnessed by adolescence researchers and practitioners. This study also demonstrates how visual data paired with survey data can be a powerful tool for simulating real-world situations and decision-making processes.
Supplemental Material
SupplementalMaterials – Supplemental material for Crowd Sourcing: Do Peer Crowd Prototypes Match Reality?
Supplemental material, SupplementalMaterials for Crowd Sourcing: Do Peer Crowd Prototypes Match Reality? by Lilla K. Pivnick, Rachel A. Gordon and Robert Crosnoe in Social Psychology Quarterly
Footnotes
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development through R01HD081022-01A1, 5R03HD096203-02, and P2CHD042849, and T32HD007081. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
1
See online Appendix A: Demographic comparison of analytic sample with peer crowd ratings versus SECCYD youth without peer crowd ratings.
2
See online Appendix B: Description of focus groups.
3
See online Appendix C: Description of crowds given to raters.
4
See online Appendix D: Demographic characteristics of undergraduate raters.
5
See online Appendix C: Description of crowds given to raters.
6
See online Appendix E: Likert scale question presented to undergraduate raters after each SECCYD video clip.
7
See online Appendix F: Reliability of crowd ratings.
8
See online Appendix G: Summary of study measures.
9
See online Appendix G: Summary of study measures.
10
See online Appendix H: Comparing p-values and effect sizes between models with imputed and non-imputed outcome variables.
11
See online Appendix I: Demographic comparison of analytic sample with academic transcript data versus those without academic transcript data.
12
See online Appendix J: Calculating predicted means and probabilities using mimrgns.
13
See online Appendix L: Pairwise comparisons between crowds hypothesized as low, middle, or high on each outcome.
14
See online Appendix K: Overall F-tests for main predictors and covariates on outcomes of interest.
15
See online Appendix M: Peer crowd pairwise comparisons for each outcome.
16
See online Appendix K: Overall F-tests for main predictors and covariates on outcomes of interest.
17
See online Appendix L: Pairwise comparisons between crowds hypothesized as low, middle, or high on each outcome.
18
See online Appendix M: Peer crowd pairwise comparisons for each outcome.
19
See online Appendix N: Summary of hypothesis testing.
20
See online Appendix M: Peer crowd pairwise comparisons for each outcome.
21
See online Appendix M: Peer crowd pairwise comparisons for each outcome.
22
See online Appendix M: Peer crowd pairwise comparisons for each outcome.
Bios
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
